MIT Report: AI Can Handle Nearly Any Undergraduate Assignment

A committee convened by MIT to study the impact of artificial intelligence on teaching stated in a report published in late August that today's AI models can reliably solve nearly any written assignment in the undergraduate curriculum, from essays to mathematical proofs. According to the report's authors, over three years the technology has driven measurable changes in campus culture.

The document, prepared by a committee made up of faculty, administrative staff, and undergraduate and graduate students, describes the situation bluntly. Generative AI, the report states, can "produce credible solutions and reasonable responses to nearly any written assignment in our undergraduate curriculum, including essays, math and science problems, proofs, and coding assignments." It is one of the most candid admissions yet from a leading American technical university that the traditional model for assessing knowledge no longer works.

What the committee found

The report's authors did not limit themselves to assessing the capabilities of the models themselves. They also documented how everyday academic life is changing. Over the three years since generative AI became widely available, attendance at faculty office hours has dropped, student participation in online course discussions has declined, and spontaneous study groups in dorms and libraries have become less common. The committee describes this as a measurable shift in campus culture, not merely a matter of isolated cheating incidents.

Rather than proposing stricter enforcement mechanisms, the report goes in the opposite direction. The committee advises against relying on software that detects AI-generated text, arguing that such tools are imperfect and can discriminate against neurodivergent students and non-native English speakers. This marks a significant departure from the approach taken by many other universities, which in recent years have widely deployed AI detectors despite their poor accuracy.

Recommendations instead of bans

Instead, the committee proposes changing how coursework is assessed: more handwritten work done in class during the early stages of projects, regular checkpoints tracking a student's progress rather than grading only the final product, and incremental feedback instead of a single final evaluation. The report also recommends that departments be allowed to update curricula on an ongoing basis, without rigid, multi-stage approval processes that take a full academic year, since the pace of change in AI capabilities is outrunning traditional university procedures.

The committee also flags the need for consistency among faculty. Instructors who use AI to create course materials or grade assignments are required to disclose this openly, to avoid hypocrisy toward students who are required to declare their own use of the same tools. The report also warns against limiting the number of top grades as a way to discourage AI use, since rationing the best grades would only increase the temptation to cut corners by relying more heavily on artificial intelligence.

Generative AI can produce credible solutions and reasonable responses to nearly any written assignment in our undergraduate curriculum, including essays, math and science problems, proofs, and coding assignments - from the report of MIT's committee on the use of AI in teaching

Other universities move toward surveillance

Not every American university is following MIT's path. The University of Chicago Law School banned first-year students from using phones and laptops in class, a move that can be read as an attempt to regain control over the teaching process through force rather than by redesigning assignments. American universities are consequently debating whether to monitor students at all while they work on assignments, and to what extent such oversight is pedagogically justified.

In Europe the question looks different, because the EU's AI Act (the bloc's artificial intelligence law) classifies systems that monitor student behavior during exams, assess the learning process, or support admissions as high-risk systems. A complete ban on emotion recognition in educational institutions has been in effect since February 2025. Some of the remaining obligations tied to classifying these systems were pushed back by sixteen months under the Digital Omnibus package, to December 2, 2027. In practice, European universities are debating not whether to monitor students, but how closely they are allowed to do so under existing law.

Relevance for Poland

For Polish universities, the MIT report is a signal that the problem will not disappear with better plagiarism detectors, since MIT's own committee considers them unreliable. Polish universities already operate under the EU AI Act, which means any student surveillance system deployed on campus must contend with the same risk classifications the report describes in the European context. The question of how to design assignments and exams in a world where AI can solve almost everything therefore applies to Polish departments as well, not just American ones.